AI Rulebook 4.4.2

A links-only patch release: the GitHub repository is now wyvernsystems/ai-rulebook. No rule text, commands, or settings changed.

Highlights

• The extension's homepage, issues, and repository links point at https://github.com/wyvernsystems/ai-rulebook. Old ai-rulebook-vscode-extension URLs redirect.
• The extension ID WyvernSystemsLLC.ai-rulebook is unchanged, so existing installs update normally.

Tests: 261 passed, 0 failed (npm run test:coverage, 96.27% line coverage).

AI Rulebook 4.4.1

A metadata-only patch release: no rule text, commands, or settings changed, so there's no reason to upgrade other than staying current.

Highlights

• The MIT license now names Wyvern Systems LLC as the copyright holder.
• The extension manifest lists Ron Picard (https://wyvernsystems.com) as the author, and the README credits Wyvern Systems.

Tests: 261 passed, 0 failed (npm run test:coverage, 96.27% line coverage).

AI Rulebook 4.4.0

AI Rulebook 4.4.0 adds a second agent mode for finishing work sooner, a one-click picker to switch modes, an on-request full code review rule, and tighter rule text that agents follow more reliably. The listing is also renamed so Cursor and Open VSX searches find it.

Highlights

• Fast parallel agent mode, off by default. Agents split a task into units that each own their files, write a shared contract, run up to five units at once (in git worktrees when they must overlap), then integrate, run the full checks, fix the combined result, and get one required review. It spends more tokens to save wall-clock time, so it cannot be on together with token-efficient subagent mode.
• AI Rulebook: Select agent mode… switches every open folder between Off, Token-efficient, and Fast parallel in one write. The sidebar has a new Agent mode row, the status bar shows the mode (AI 7/7 · parallel), and all three warn if a hand edit or merge leaves both modes on. The sidebar toolbar's four mode entries are replaced by this one picker; the enable and disable commands stay in the command palette.
• New Review rule (review.mdc). Ask an agent for a full code review and it works through every unpushed change in passes for bugs, security, dead code, duplication, improvements, tests, and docs, fixing as it goes. It reports licensing concerns without fixing them, runs your tests, lint, and type checks, reports by pass, and then commits and pushes, never force-pushing. It never starts on its own.
• Agents fix what they find. The Scope rule now has agents fix small, local, check-verified defects they come across outside the task and list them under Also fixed, instead of only reporting them. Fixes that would change interfaces or behavior, need a migration, or need your decision are still only reported.
• Sharper rule text. One instruction per bullet, the most often broken instruction first, lines models already follow removed, and a fixed report shape for every task: Changed, Also fixed, Checks, Not done. The Docs rule asks for dense documentation written for the reader who acts on it.
• Automatic refresh. The sidebar and status bar update when AGENTS.md changes on disk (pull, branch switch, hand edit).
• Easier to find. The listing is now AI Rulebook: AGENTS.md Rules for Coding Agents, with keywords for AGENTS.md, CLAUDE.md, and the supported tools. In Cursor you can also search for WyvernSystemsLLC.ai-rulebook.

Upgrading

Run AI Rulebook: Install / update rule pack (or accept the update prompt) to refresh the rule text in AGENTS.md. The new Review rule is added enabled; turn it off in the sidebar if you don't want it. Rules you disabled stay disabled. The extension ID is unchanged, so installs update normally.

Verification

npm run lint is clean. npm run test:coverage: 261 tests, 261 passing, 96.27% line, 89.03% branch, and 86.43% function coverage.

AI Rulebook 4.3.0

AI Rulebook 4.3.0 adds an optional mode that cuts agent cost: the expensive session model plans and judges, while searching, editing, and check runs go to subagents on cheaper models. It lives entirely in AGENTS.md, so it adds no per-tool folders or settings to your repository.

Highlights

• Token-efficient subagent mode, off by default. AI Rulebook: Enable token-efficient subagent mode adds its own managed block to AGENTS.md in every open folder; AI Rulebook: Disable token-efficient subagent mode removes it. The six-rule pack, Enable all rules, and Remove rule pack never turn it on or remove it.
• Works across tools. The text assigns work to fast, standard, and deep model tiers instead of one vendor's model names, so Claude Code, ChatGPT / Codex, Cursor, opencode, and GitHub Copilot agents can each apply it with their own models. Tools without subagents are told to skip it.
• Substitutes when a model is unavailable. Fast falls back to standard and then the session model, standard to the session model, and deep to the session model or standard at high effort. The agent never picks a model more expensive than the session model, and tells you once which substitute it chose.
• No retry loops. When a delegate's edit is wrong or its check fails, the orchestrator fixes it itself. An incomplete spec is resent once, and the same change never gets a third delegate.
• Accuracy checks kept. High-risk diffs (concurrency, security, auth, credentials, persistence, migrations) get one read-only review on the strongest model after the checks pass, and the orchestrator reads the final diff before reporting.
• Show rule pack status reports whether the mode is on.

Upgrading

No action is needed. Existing AGENTS.md rule packs are unchanged, and the new mode stays off until you run its enable command.

Verification

npm run lint is clean. npm run test:coverage: 201 tests, 201 passing, 96.30% line, 87.90% branch, and 85.23% function coverage.

AI Rulebook 4.2.0

AI Rulebook 4.2.0

AI Rulebook 4.2.0 refines the engineering rules and makes filesystem failures easier to diagnose while preserving all six rules and their toggles.

This release includes the changes prepared for 4.1.0, which was tagged but not published as a GitHub release.

Highlights

• Clearer task-completion guidance, protection for existing work, behavior-focused tests, and final verification.
• The Tests rule recommends a relevant linter when absent and allows setup when requested or approved.
• Simpler documentation guidance and clearer dependency, compatibility, and error-handling requirements.
• Unreadable or malformed rule files show Unable to read in the sidebar instead of appearing disabled.
• Project detection and legacy cleanup report filesystem failures; wrapped errors preserve their original cause.
• Recommended ESLint and typescript-eslint checks now run locally and in CI.

Upgrading

1. Install ai-rulebook-4.2.0.vsix using your editor's Install from VSIX action.
2. Run AI Rulebook: Install / update rule pack to refresh the installed guidance. Existing enabled and disabled states are preserved.

Downloads

• ai-rulebook-4.2.0.vsix: installable extension package.
• AGENTS.md: standalone pack. Preserve existing project instructions when copying the managed block, include both outer markers, and replace the generic test-command wording. For Claude Code, append @AGENTS.md to CLAUDE.md without overwriting its content.

Validation

• npm ci completed using the lockfile.
• npm run lint: passed with zero warnings.
• npm test: 181 passed, zero failed or skipped; includes TypeScript compilation and bundled-rule verification.
• npm run test:coverage: 96.28% lines, 87.87% branches, 85.09% functions across all reported files, including the test mock, on Node.js 26.7.0.
• npm run package: succeeded; VSIX version, compiled code, bundled rules, and linter configuration exclusion verified.
• GitHub CI passed for commit 02e00dd: lint on Node 24 and tests on Node 18 and 20. Existing v4 Actions emit a Node 20 runtime deprecation warning.
• Production-dependency audit: zero vulnerabilities.
• Nine existing development-dependency vulnerabilities remain (two moderate, seven high); no existing dependency versions were changed when adding ESLint tooling.

AI Rulebook 4.0.0

AI Rulebook 4.0.0

AI Rulebook now installs one shared rule pack in your project's AGENTS.md, with a CLAUDE.md import for Claude Code. This major release replaces the per-tool mirror layout and requires migration for existing installs.

Highlights

• Six focused rules in one managed block, with sidebar checkboxes that remove disabled rule text and restore it when enabled.
• Install/update preserves the block's enabled states and applies to every open workspace folder.
• Automatic installation checks for AI-agent use; projects without evidence receive a one-time hint.
• Rule status and navigation now point to AGENTS.md.
• Concurrent rule mutations within one extension host are serialized to prevent lost updates and malformed files.
• Claude import cleanup preserves fenced examples and existing prose.
• Standalone distribution is a single AGENTS.md; the VSIX excludes old ZIP archives and repository agent guidance.
• Updated documentation, removed obsolete code and local rule copies, and added regression tests.

Upgrading

1. Install ai-rulebook-4.0.0.vsix using your editor's Install from VSIX action.
2. Run AI Rulebook: Install / update rule pack to create or refresh the managed block in AGENTS.md and ensure the Claude import.
3. Run AI Rulebook: Remove legacy per-tool rule folders… to remove old mirrors.
4. Remove .opencode/rules/ai-rules/*.md from your opencode configuration's instructions if present; the cleanup command does not edit that configuration.
5. Reapply any rules disabled in the old layout. Legacy disabled states are not imported; a new block starts with all six rules enabled.
6. Commit AGENTS.md and CLAUDE.md to share the rules with your team.

Per-tool sync/remove commands, reset-to-defaults, Explorer rule coloring, and per-tool settings have been removed. Install, remove, and legacy cleanup apply to all workspace folders; rule toggles and status use the first folder. Copilot support varies by feature and settings; see the README's per-tool notes.

Downloads

• ai-rulebook-4.0.0.vsix: installable extension package.
• AGENTS.md: standalone pack. Preserve existing project instructions when copying the managed block, include both outer markers, and replace the generic test-command wording. For Claude Code, append @AGENTS.md to CLAUDE.md without overwriting its content.

Validation

• npm ci completed using the lockfile.
• npm test: 151 passed, zero failed.
• GitHub CI passed on Node.js 18 and 20 for release commit 42693c4.
• npm run test:coverage: 94.44% lines, 84.41% branches, 84.78% functions across all reported files, including the test mock, on Node.js 26.7.0.
• npm run package succeeded; the VSIX version, exclusions, compiled code, and bundled rules were verified against the release checkout.
• The production-dependency audit reports zero vulnerabilities. npm reports nine existing development-dependency vulnerabilities (two moderate, seven high); dependency upgrades are not included in this release.

AI Rulebook 3.1.0

AI Rulebook 3.1.0 brings the rule pack to two more tools — Windsurf and
GitHub Copilot — and to two more distribution channels: the extension is now
live on Open VSX,
so Cursor, Windsurf, VSCodium, and other VS Code forks can install it
straight from their Extensions panel.

Highlights

• Windsurf support. The rule pack mirrors to .windsurf/rules/ai-rules/
  with trigger: frontmatter, auto-synced when the workspace shows evidence
  of Windsurf (a .windsurf/ folder or .windsurfrules file). New sync and
  remove commands, Explorer color-coding, and an
  aiRules.autoSyncWindsurfWhenInstalled setting.
• GitHub Copilot support. The rule pack mirrors to
  .github/instructions/ai-rules/ as *.instructions.md files with
  applyTo: frontmatter, auto-synced when the workspace uses Copilot custom
  instructions. New sync and remove commands and an
  aiRules.autoSyncCopilotWhenInstalled setting.
• Open VSX publishing. Releases now also publish to Open VSX, the
  registry used by Cursor, Windsurf, VSCodium, code-server, and Eclipse
  Theia. Same .vsix as the GitHub release, so channels cannot diverge.
• CI. Every push and pull request now runs the full test suite on
  Node 18 and 20.
• README distribution and compatibility docs. New sections describe the
  three distribution channels and separate the editors the extension runs in
  from the tools it writes rules for.

Standalone rule packs for Windsurf and Copilot are attached below alongside
the existing four, for anyone who wants the rules without the extension.

Quality: 223 tests, all passing; coverage 90.69% lines / 81.92% branches /
82.80% functions.

Upgrading

No action needed. Existing workspaces pick up the Windsurf and Copilot
mirrors on the next install, reset, or sync — or automatically when the
workspace shows evidence of those tools.

AI Rulebook 3.0.0

This release rewrites the rule pack around one question: can an agent
actually follow each rule, deterministically, mid-task? The docs rule is
rebuilt around the wider set of conventional documentation files and
Spec Kit-style requirements, and the other rules trade vague judgment
calls for checkable thresholds. It's a major version because rule
guidance changed shape: two docs triggers were removed or replaced
(docs/TESTING.md folds into CONTRIBUTING.md; single-file requirements
give way to GitHub Spec Kit specs), so teams relying on the old docs
behavior should review before upgrading.

Highlights

• Rebuilt docs rule. Covers README.md, CHANGELOG.md (Keep a
  Changelog form, user-facing entries), AGENTS.md, CONTRIBUTING.md,
  Spec Kit feature specs (specs/<NNN-feature>/spec.md),
  docs/ARCHITECTURE.md, docs/RELEASING.md, docs/DEPLOY.md, and
  docs/decisions/ ADRs — each behind an update trigger that fires only
  when the file exists. Agents scan the trigger list once before
  reporting a task done.
• No speculative docs. New documentation is created only by
  graduating a README section that outgrew ~50 lines, or on a
  first-occurrence trigger; specs/ only when the project uses Spec Kit
  or the user asks; the first ADR only after confirming with the user.
• git rule covers requested commits. Stage only the task's files,
  imperative subject with a why — and never force-push, rebase, amend,
  or reset away pushed history unless the user names the operation.
• code rule got checkable. Prefer the standard library over new
  dependencies, "unmaintained" now means archived, deprecated, or no
  release in roughly two years, and input validation is scoped to trust
  boundaries.
• tests rule scoped. Test-first applies to behavior changes (so
  refactors can't silently skip it), and the no-convention fallback is
  deterministic: ask when the user is available, colocate when working
  autonomously.
• scope rule closes the drive-by loophole. An unrelated bug or
  dead code you notice gets reported, not fixed.

185 tests pass with 90.22% line / 81.98% branch coverage.

Upgrading

After updating the extension, run AI Rulebook: Sync rule pack to all
formats to refresh installed workspaces with the new rule text while
keeping each rule's on / off state (or Install / update rule pack to
start fresh with everything enabled). If your project relied on the old
docs/TESTING.md or docs/REQUIREMENTS.md triggers: testing docs now
belong in CONTRIBUTING.md, and an existing docs/REQUIREMENTS.md is
still honored by the rule's match-your-convention clause.

AI Rulebook 2.10.0

This release fixes the manual sync and remove commands. The headline bug: Sync
rule pack to all formats silently re-enabled every rule you had turned off, in
all four formats at once. If you use the sidebar toggles, upgrade.

Highlights

• Sync commands preserve rule state. Sync rule pack to all formats and
  Sync rule pack to Cursor refreshed .cursor/rules/ai-rules/ from the
  bundle before generating the mirrors, which deleted every .disabled marker
  and turned every rule back on across Cursor, Cline, opencode, and Claude
  Code. The Sync commands now push exactly what the sidebar shows.
  Install / update rule pack and Reset rule pack to defaults still start
  from the bundled defaults, where every rule is enabled.
• Sync to all formats covers every workspace folder. It wrote only the
  first one, while the per-format sync commands wrote all of them. The
  .cursor/ install stays scoped to the first folder, as documented.
• Remove commands clear every workspace folder. In a multi-root workspace
  they deleted the first folder's rule packs and reported that everything was
  removed. The confirmation dialog now names the folder count.
• The status bar stops going stale. The Cline, opencode, and Claude Code
  sync and remove commands never refreshed it — so clicking the status bar to
  sync opencode left its own indicator unchanged.
• Rule toggles no longer report a false success. Without a
  .cursor/rules/ai-rules/ folder, the enable / disable commands and the
  sidebar checkboxes did nothing and still said they succeeded. They now point
  at Install / update rule pack.
• Two dead command-surface bugs fixed. Open rule file is hidden from the
  command palette, where it had no rule argument to act on; Hide rule colors
  no longer aims a folder-scoped update at a resource-less configuration,
  which VS Code rejects.
• A command reference in the README. All 20 commands in one table, with
  what each writes and which workspace folders it touches.

Upgrading

No action needed. Installed workspaces keep their rule files and their on/off
state.

If you previously ran Sync rule pack to all formats and found rules you had
disabled were active again, re-disable them once from the sidebar — the sync
will respect that from now on.

Verified with 185 tests: 90.22% line, 81.98% branch, 82.15% function coverage.

AI Rulebook 2.9.0

This release rounds out multi-tool support: mirroring to Cline, opencode, and Claude Code now works correctly across every folder in a multi-root workspace, opencode gains a slash command for inspecting rule state without leaving the terminal, and a status bar item surfaces rule/sync state at a glance.

Highlights

• Status bar item. Shows the enabled-rule count for the first workspace folder plus the opencode config sync state, and clicks through to Sync rule pack to opencode.
• Multi-root workspace support. Cline, opencode, and Claude Code mirroring — and their manual sync commands — now run independently in every open workspace folder, judged on that folder's own evidence and Cursor rule state. (The .cursor/rules/ai-rules/ install itself stays scoped to the first folder.)
• Explorer coloring for Cline mirrors. .clinerules/ai-rules/ files are now tinted green / red alongside the Cursor, opencode, and Claude Code mirrors.
• opencode /ai-rulebook command. Syncing to opencode now writes .opencode/command/ai-rulebook.md, a slash command that lists active and disabled rules from inside opencode.
• Fix: auto-sync now warns when the project's opencode config can't be parsed — previously only the manual sync command surfaced that failure.
• Fix: Install / update rule pack no longer stops syncing opencode/Claude Code after the first matching folder in a multi-root workspace.

168 tests passing, 86.74% line coverage.